Lightweight Adaptive and Mixed Concurrency Control

Monday, November 13, 2017 -
12:00pm to 1:00pm
2310 CS

Speaker Name: 

Aaron J. Elmore

Speaker Institution: 

University of Chicago

Cookies: 

No

Description: 

Title: Lightweight Adaptive and Mixed Concurrency Control

Abstract: New concurrency control protocols focus on enabling high throughput for large main-memory and many-core systems. These protocols are largely optimized for specific workloads, but due to unknown and shifting access patterns one specific algorithm cannot fit all varied workloads. Thus, it is desirable to choose the right concurrency control protocols for a given workload. To address this issue we present ACC, a database system that dynamically applies a concurrency control algorithm to partitions of the database according to workload characteristics and supports transaction execution across mixed concurrency control protocols. This talk will introduce a general framework for mixing multiple protocols without modifying them or introducing any overhead of coordinating conflicts across protocols, a set of simple and efficient features for selecting an ideal protocol, a method for reconfiguring protocols without stopping execution, and a prototype system. Our experiments show compared with static protocols, ACC achieves throughput by up to 9.9x, 2.2x, and 1.9x over a partitioned single-threaded concurrency control, OCC, and 2PL respectively.

Bio: Aaron J. Elmore is an Assistant Professor in the Department of Computer Science, and the College of the University of Chicago. Aaron was previously a Postdoctoral Associate at MIT working with Mike Stonebraker on elastic and multitenant database systems, and Sam Madden on the DataHub project. Aaron's thesis on Elasticity Primitives for Database-as-a-Service was completed at the University of California, Santa Barbara under the supervision of Divy Agrawal and Amr El Abbadi. His research interests include elastic and adaptive database systems, collaborative data analytics, tools for data lake management, and data intensive systems.